Test prediction inputs and explainability

Read model information and make sure the deployed model expects the modality and feature structure you plan to submit.

For compatible vision and medical models, drag or choose a supported image, TIFF, DICOM, or NIfTI input. Review the selected file before submission.

For tabular models, complete the displayed feature fields using the same units and categorical conventions used during training.

For a language model, enter the prompt and set the exposed maximum-token and temperature controls. Start conservatively and avoid sensitive prompt content.

If an inactive-model confirmation appears, cancel unless you are authorized to activate the deployment. Otherwise confirm and wait for the visible processing state to finish.

Start with the task-aware summary and target legend. Confirm that the displayed result belongs to the input and deployment version you just used.

Use the raw-response view or download for the complete structured result. Treat it as model output, not a clinical or operational decision by itself.

For media tasks, inspect the protected output, overlays, masks, boxes, classes, scores, or generated media available for that task.

For compatible panoptic segmentation, use the legend and overlay to distinguish semantic classes and individual regions, then cross-check the raw response.

For tabular models, inspect available SHAP waterfall or force views. For image models, inspect the configured Grad-CAM or Integrated Gradients output. Explanations indicate model influence, not correctness.

Use history paging to find the test and select View to reopen its saved result. Confirm its state before using it in a report.

Choose the cURL or Python example that matches your server environment.

Do not paste a real API key, private input, or production identifier into documentation, screenshots, issue reports, or client-side code.

Store the API key in a server-side secret manager and supply the required authentication header from trusted code.

Use the Headless API documentation for authentication, request fields, asynchronous status, protected artifacts, and endpoint-specific error handling.